Empirical Analysis
To test Between-sector Job Polarization across EU member states I am intending to adopt a model developed in the paper, Goos et al. (2015), where they look at the regression model outlined below.
DeltaYjct=bo+b1LnCapjc+b2LnCapjc^2+ejct
Where:
DeltaYjct = % difference between t and t-5 in the share of workers employed in each industry j in country c
LnCapjc = ICT capital intensity index of industry j in country c
Ejct= error term
We are interested in finding a U shape relationship between DeltaYj and LnCap
Hence the hypothesis to test the presence of Job Polarization is as follows: b1>0 and b2>0
For the reallocation of skilled workers (who are defined to be workers with tertiary education) a similar regression model is used, only DeltaYjct changes to % difference between t and t-5 in the share of skilled workers employed in industry j in country c
The hypothesis we are testing is b1>0 and b2>0 be insignificant
And when we set the value of b2 equal 0, b1 estimated coefficient should be positive and significant
For the reallocation of unskilled workers (who are defined to be workers with primary and secondary education), we again use the same regression model, however in this case DeltaYjct indicates the difference in share of unskilled workers
Moreover, the hypothesis is that the model will predict more positive change in the employment share for the least compared to middling ICT capital intense sectors.
All data is gathered from EUKLEMS 2017 database:
To calculate ICT capital intensity index, I have developed the following formula using 2014 dataset:
ICTCapjc=[(Total Capital Compensations)x(Nominal Gross fixed Capital Formation of ICT capital/Total Nominal Gross fixed Capital formation)]/(Gross Output)
In terms of indexes used in the EUKLEMS database:
ICTCapjc= [(Cap)x(I_IT+I_CT+I_Soft_DB)/(I_GFCF)]/(GO)
However, for some countries the Nominal Gross fixed Capital Formation of ICT capital was missing hence I have used slightly different formula. Instead of aggregating ICT capital I have deducted non-ICT capital, from Total Nominal Gross fixed Capital formation. As this can be seen in the excel spreadsheet, which contains all of the formulas for the indexes.
Due to missing data for some of the industries only the following industries would be used in the empirical model:
By using the ranking of GDP per capita I have grouped the countries of my interest in two groups.
Group 1: (Denmark, Sweden, Finland, Austria, Netherlands, Germany, France, U.K., Italy, Spain)
Group 2: (Slovenia, Portugal, Czech R., Estonia, Slovakia, Latvia, Poland, Hungary)
For the first part of the empirical analysis where we test the relationship of ICT capital intensity on Total hours worked by person engaged if we used the time period of 1995-2014, only 4 countries from Group 2 would have sufficient data (Slovenia, Portugal, Czech R., and Slovakia) as for Estonia, Latvia and Poland the data is available since 2000.
For the second part of empirical analysis, EUKLEMS has missing data for all countries during 2006 and 2007. Hence it is possible to either perform the analysis on countries for the period 2008-20014 (and data is complete for all countries)
Or test for 1995-2005 using earlier EUKLEMS data sources. However, Sweden and France from first group and Portugal, Estonia, Latvia from the second group should be dropped.
Given the data perhaps for the analysis of skilled vs unskilled employers 2008-2014 years would be more appropriate??
On the other side in the original paper, DeltaYjct, the % change in the share of the workers have been done in differences between 5-year time period, and it is done to take into account of business cycle fluctautions in the economies. But given the data perhaps a shorter time period difference should be used if the model does not suffer too much from the change?
And lastly the data are given in EUKLEMS on labour Input also contains information on the gender of workers, does that mean that the model would benefit from taking into account the gender of the workers as well or it would not change our findings?
As in the original paper the results are presented on Kernel Plots of average annual employemtn share changes by a ranking of sectors according to their ICT capital intensity in 2014 averaged across countries.